DataGrid develops AI to generate whole-body images of nonexistent people
An artificial intelligence algorithm from Japanese company DataGrid creates high-resolution whole-body images of non-existent people.
DataGrid uses a generative adversarial network (GAN) to automatically generate 1024 x 1024 images of models against a plain white background. The technology is the product of research and development of “whole body generation” and “motion generation.” The company says while similar techniques have been used to generate facial images, its high-precision whole-body generation has no precedent, according to a Google translation of a company announcement.
Even those faces often had obvious flaws, like asymmetrical ears, Futurism reports. The faces created from scratch using GANs at Thispersondoesnotexist.com seem fairly consistent in their realism, however, raising concerns about the potential of the technology to be used for deepfakes.
The company plans to further improve the accuracy of its system, and conduct demonstrations with advertising and apparel companies to develop an operational product.